Don't let replica-based cache updates race real writes

What does this MR do and why?

Don't let replica-based cache updates race real writes.

... and split mention calculation from store so we can minimise the time spent with a row lock.

See #628134 (comment 3945257084) for background on the new query, and that WI in general for what this work is driving towards.

Sample query

UPDATE "notes" SET "note_html" = '<p data-sourcepos="1:1-1:21" dir="auto">Note mentioning <a href="/root" title="Administrator" class="gfm gfm-project_member js-user-link" data-user="1" data-original="@root" data-container="body" data-placement="top" data-reference-type="user">@root</a></p>', "cached_markdown_version" = 2228224, "namespace_id" = NULL WHERE "notes"."id" = 33 AND (COALESCE(cached_markdown_version, 0) <= 2228224) AND "notes"."note" = 'Note mentioning @root'

Explain: https://console.postgres.ai/gitlab/projects/gitlab-production-main/sessions/62312/commands/167454

 Update on public.notes  (cost=0.58..3.60 rows=0 width=0) (actual time=8.906..8.906 rows=0 loops=1)
   Buffers: shared hit=3 read=5
   I/O Timings: read=8.839 write=0.000
   ->  Index Scan using notes_pkey on public.notes  (cost=0.58..3.60 rows=1 width=50) (actual time=8.904..8.904 rows=0 loops=1)
         Index Cond: (notes.id = 33)
         Filter: ((COALESCE(notes.cached_markdown_version, 0) <= 2228224) AND (notes.note = 'Note mentioning @root'::text))
         Rows Removed by Filter: 1
         Buffers: shared hit=3 read=5
         I/O Timings: read=8.839 write=0.000
Settings: random_page_cost = '1.5', seq_page_cost = '4', effective_cache_size = '472585MB', work_mem = '230MB', jit = 'off'

Statistics:
Time: 26.653 ms
  - planning: 17.627 ms
  - execution: 9.026 ms
    - I/O read: 8.839 ms
    - I/O write: 0.000 ms

Shared buffers:
  - hits: 3 (~24.00 KiB) from the buffer pool
  - reads: 5 (~40.00 KiB) from the OS file cache, including disk I/O
  - dirtied: 0
  - writes: 0

MR acceptance checklist

Evaluate this MR against the MR acceptance checklist. It helps you analyze changes to reduce risks in quality, performance, reliability, security, and maintainability.